Choose a model, set your usage, and see what your generative AI workload costs in GenAI tokens, Capacity Units and euros - with every step of the formula shown.
€1 is an example. Use your contract rate.
€60.11
€721.37 per year at the same usage
Where the cost goes
GenAI tokens = [(3,500 ÷ 1,000 × 0.00024) + (300 ÷ 1,000 × 0.00141)] × 25,000 = 31.575
CU = 31.575 × 1.90385 = 60.1141
EUR = 60.1141 × €1.00 = €60.11
GPT-5 Mini rates, version 2025-08-07, from SAP Note 3437766.
Same workload, monthly cost per model. Cheapest first.
SAP does not bill generative AI hub by the model provider’s token price. Usage is converted twice: first from model tokens to GenAI tokens, using a rate per model, then from GenAI tokens to Capacity Units, the currency of SAP BTP consumption.
Input and output tokens per request, divided by 1,000, times the model’s input and output conversion rates, times requests per month.
Monthly GenAI tokens times the CU conversion factor. The default factor comes from SAP’s worked metering example.
Capacity Units times your price per CU. €1 is a placeholder: enter the rate from your BTP contract.
Tokens, requests and Capacity Units per agent, project and model - so you fix the one agent that burns half the budget.
Compare real cost per request across models, and move routine steps to a smaller one without guessing.
MCP servers for SAP return only the fields a task needs, instead of whole documents pasted into the prompt.
Token threshold alerts and a month-end CU projection, before the invoice arrives.
A calculator answers “what will this cost?” before you build. Once agents run, the question becomes “what did it cost, and who used it?”. The Boidra Platform tracks tokens, requests and Capacity Units per agent and project, so FinOps can validate real spend against this estimate. If you are still choosing an architecture, see GenAI applications for SAP.
Model usage is metered in GenAI tokens. Each model has its own conversion rate for input and output tokens, published in SAP Note 3437766. Monthly GenAI tokens are converted to Capacity Units (CU), and you pay the CU price in your SAP BTP contract.
SAP’s billing unit for generative AI hub. It is not the same as a model token: the rate per 1,000 model tokens differs by model and by direction, so the same prompt costs a different number of GenAI tokens on GPT, Claude, Gemini or Mistral.
For every model in the note, the output conversion rate is higher than the input rate, often four to six times. Long answers therefore drive cost faster than long prompts, which is why the calculator shows the split.
It is a basic calculation of standard input and output model tokens only. It does not include orchestration, prompt optimization, evaluations, grounding queries and storage, translation, embeddings or reranking, which SAP’s official AI Core Cost Calculator prices. Cached tokens, batch discounts and taxes are not included either.
No. It is an independent planning tool built by Boidra, using rates published in SAP Note 3437766. For a binding figure, use SAP’s own estimator and your contract.
An estimate tells you what to expect; the Boidra Platform tells you what happened. It tracks tokens, requests and Capacity Units per agent and project, alerts on token thresholds, and projects month-end spend, so FinOps can compare real spend against the plan.
Conversion rates come from SAP Note 3437766, Availability of Generative AI Models, version 155, released 28 July 2026 (SAP for Me login required). The GPT 5.6 and Gemini 3.5 Flash Lite rows come from a later version of the note. Availability and rates change, and some listed models have retirement dates. The CU factor comes from SAP’s metering and pricing guide.
An independent planning tool, not an official SAP calculator. For binding figures use SAP’s AI Core calculator and your contract.
Tell us what you’re trying to automate in SAP. We’ll tell you - honestly - whether an agent is the right tool, and how we’d start.
